Documentation: tuples vs lists

[neighthan]

Issue: different behavior for lists vs tuples

I've been confused multiple times because of how holoviews handles tuples vs lists. Here are a couple of examples:

# imports / dummy data
import numpy as np
import holoviews as hv
hv.extension("bokeh")

x = np.arange(10)
y = 2 * x + 3
c = np.linspace(0, 1, len(x))

As input data to a plot

hv.Scatter((x, y, c))

• this generates the plot I expect

hv.Scatter([x, y, c])

• this only plots 3 points, none of which are "correct". From looking at hv.Scatter([x, y, c]).data, it assumes that the first two entries in each array are, respectively, the (x, y) coordinates of the points.

For vdims

So, I've learned that, for how I format my data, I should use tuples instead of lists. Then I try to rename my vdims:

hv.Scatter((x, y, c), kdims="h", vdims=("w", "c"))

• This gives me a plot where the y-axis is labelled "c". Looking at hv.Scatter((x, y, c), kdims="h", vdims=("w", "c")).data shows I just have h and w stored for each point; c isn't present at all. I'm not sure how using a tuple here is being interpreted or what the proper way to use one would be.

hv.Scatter((x, y, c), kdims="h", vdims=["w", "c"])

• This generates the plot I expect with the x-axis labelled "h", y-axis labelled "w", and (h, w, c) in the data.

Solution: documentation?

For me, it'd be nice if, e.g., hv.Scatter((x, y)) and hv.Scatter([x, y]) were equivalent, but I assume some other people are depending on the current behavior of [x, y], so this is just a request for clearer documentation. I don't know where would be the best place to put that information, though (the FAQ? unless I'm the only one for whom this issue is "frequent" enough). This information is probably already available somewhere, but I'm not sure how I'd easily find it when, e.g., I don't understand why my vdims aren't working how I expect..

[neighthan]

Ah, the FAQ does address what tuples in vdims are used for; I didn't see it since I was searching for the word vdims. If one was passing in a dataframe, then a tuple of (column_name, plot_axis_name) could be given.

[jlstevens]

Thanks for reporting this. Maybe we should add the word vdims to the FAQ so this behavior is more searchable..